race
PulseAugur coverage of race — every cluster mentioning race across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
New Verdict-MQ model efficiently answers multiple questions from single state encoding
Researchers have developed Verdict-MQ, a novel encoder model designed to efficiently answer multiple questions about a single piece of information. Unlike traditional cross-encoders that process each question separately…
-
LLMs show bias toward majority groups when prompted with demographics
A new study has revealed that large language models (LLMs) do not act as neutral judges when prompted with demographic information. Instead, models without any demographic conditioning tend to align with the judgments o…
-
New D2C-Routing method improves AI-generated text detection
Researchers have developed a new method called D2C-Routing for detecting AI-generated text, particularly in cases of mixed origin where content and expression sources differ. This approach frames the problem as dimensio…
-
New RACE framework offers scalable analysis of LLM neuron consistency
Researchers have introduced RACE (Residual Alignment for Consistency Estimation), a new statistical framework designed to evaluate the functional consistency of neurons within Transformer models. This method aims to add…
-
AI continues to struggle with race and gender bias
AI systems continue to exhibit significant biases related to race and gender, a problem that persists despite ongoing development. These biases manifest in various AI applications, highlighting a critical challenge for …
-
Research paper analyzes compute allocation for RL post-training
A new research paper explores how to best allocate limited compute resources for reinforcement learning (RL) post-training of foundation models. The study introduces a FLOP-accounting framework to analyze the trade-offs…
-
New ML framework tackles intersectional bias with coverage constraints
Researchers have developed a new framework to mitigate bias in machine learning models, particularly for individuals at the intersection of multiple sensitive attributes like race and gender. This approach incorporates …